Data cleaning method and device for target detection of unmanned aerial vehicle group

Abstract

The invention relates to the technical field of data processing, in particular to a data cleaning method and device for target detection of an unmanned aerial vehicle group. The method comprises the steps that radar echo signals generated by an unmanned aerial vehicle under the influence of an interference source in a target area are collected, an initial feature engineering set is formed, and the radar echo signals comprise target echo signals and interferent echo signals; the initial feature engineering set is processed through a data cleaning network framework, a target data set is generated, the data cleaning network framework comprises a distance gate module and a feature classification module, the input initial feature engineering set is processed through a serial or parallel network construction method, and the target data set is generated. And generating a target data set of the filtered clutter trace points. According to the data cleaning method for target detection of the unmanned aerial vehicle group provided by the embodiment of the invention, clutter removal processing is performed on the collected initial feature engineering set through the data cleaning net framework, the clutter removal efficiency can be improved, and a high-quality target data set can be generated.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a data cleaning method and device for drone group target detection. Background Art

[0002] At present, low-altitude UAVs are growing rapidly in civil needs and scenarios, and are also an important means of transportation in the low-altitude economy. In order to effectively manage low-altitude airspace, it is urgent to build technical methods and equipment systems that can effectively detect UAV formations, and discover, identify and track UAV group targets in real time.

[0003] In urban low-altitude detection applications, some typical scenarios require the generation of high-quality detection data. For example, when applying Bayesian theory and methods, the prediction values ​​generated by the prediction equation are difficult to obtain high-quality prediction sequence values ​​due to the limitations of model quality; due to the data association and clutter problems of the drone formation, the measurement values ​​are also of poor quality. By generating higher-quality data as a correction and supplement to the measurement, better measurement values ​​can be obtained; in complex low-altitude environments, the missing data of multiple detection cycles leads to discontinuous tracks, and high-quality track completion data needs to be generated.

[0004] In view of this, how to provide a data cleaning method to efficiently remove clutter and obtain high-quality data samples has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide a data cleaning method for drone swarm target detection, a data cleaning device for drone swarm target detection, an electronic device and a computer storage medium, which are used to solve the current problems of low clutter removal efficiency and low data sample quality.

[0006] In a first aspect of an embodiment of the present application, a data cleaning method for drone group target detection is provided, comprising:

[0007] Collect radar echo signals generated by UAVs in the target area under the influence of interference sources to form an initial feature engineering set, wherein the radar echo signals include target echo signals and interference echo signals;

[0008] The initial feature engineering set is processed by a data cleaning network framework to generate a target data set, wherein the data cleaning network framework includes a range gate module and a feature classification module, and processes the input initial feature engineering set by adopting a series or parallel network construction method to generate a target data set that filters clutter points.

[0009] In a second aspect of an embodiment of the present application, a data cleaning device for drone group target detection is provided, comprising:

[0010] A collection module is configured to collect radar echo signals generated by the UAV in the target area under the influence of the interference source to form an initial feature engineering set, wherein the radar echo signals include target echo signals and interference echo signals;

[0011] The processing module is configured to process the initial feature engineering set through a data cleaning network framework to generate a target data set, wherein the data cleaning network framework includes a range gate module and a feature classification module, and processes the input initial feature engineering set by adopting a series or parallel network construction method to generate a target data set that filters out clutter points.

[0012] According to a third aspect of an embodiment of the present application, a computing device is provided, including:

[0013] Memory and processor;

[0014] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the data cleaning method for drone swarm target detection are implemented.

[0015] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the data cleaning method for drone swarm target detection are implemented.

[0016] The present application provides a data cleaning method for drone swarm target detection, comprising: first, collecting radar echo signals generated by drones in a target area under the influence of interference sources to form an initial feature engineering set, wherein the radar echo signals include target echo signals and interference echo signals; then, processing the initial feature engineering set through a data cleaning network framework to generate a target data set, wherein the data cleaning network framework includes a range gate module and a feature classification module, and processes the input initial feature engineering set by adopting a series or parallel network construction method to generate a target data set that filters clutter points.

[0017] By applying the data cleaning method for drone swarm target detection provided in the embodiment of the present application, the collected initial feature engineering set is processed for clutter removal through a data cleaning network framework, which can improve the clutter removal efficiency and generate a high-quality target data set.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, which can be implemented in accordance with the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0020] Figure 1 A flow chart of a data cleaning method for drone swarm target detection provided in an embodiment of the present application;

[0021] Figure 2 A schematic diagram of a serial data cleaning network structure for drone swarm target detection provided in an embodiment of the present application;

[0022] Figure 3 A schematic diagram of a parallel data cleaning network structure for drone swarm target detection provided in an embodiment of the present application;

[0023] Figure 4 A schematic diagram of the structure of a data cleaning device for drone swarm target detection provided in an embodiment of the present application;

[0024] Figure 5 A structural block diagram of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0026] The activities of drone swarms are characterized by high mobility and changeable routes. In addition, there are many interference sources in the low-altitude urban environment, such as swaying branches, flying birds, and building obstructions. This makes the low-altitude multi-target detection data contain a lot of clutter and false wave components, which brings huge challenges to the technical task of point-track fusion. Due to the complexity of the low-altitude environment, it is very common for perception equipment to report intermittent target tracks, especially when track points lose multiple detection cycles. This generally leads to interruption of the target track; when the target track information reappears, a new batch number will be generated, resulting in the inability to grasp the overall operation process of the target, which is not conducive to the judgment of the target's tactical intentions. The large amount of clutter and high randomness, as well as the difficulty in associating multi-target data, have become key issues that need to be urgently addressed in multi-target tracking technology.

[0027] In response to the needs of multi-target detection in low-altitude complex environments under civil radar detection means, aiming at a series of correlation problems such as the large number of false points in the measurement data, complex correlation of measurement data, and difficulty in generating formation track fusion, this paper studies the combined application of artificial neural networks and traditional data association methods, and constructs a method path for cleaning and generating target track data for single UAVs or UAV formations through algorithm research and system design for data generation and clutter removal.

[0028] See also Figure 1 , Figure 1 A flow chart of a data cleaning method for drone group target detection provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the specific steps include the following steps.

[0029] Step S102: collecting radar echo signals generated by the UAV in the target area under the influence of the interference source to form an initial feature engineering set, wherein the radar echo signals include target echo signals and interference echo signals.

[0030] Step S104: The initial feature engineering set is processed through a data cleaning network framework to generate a target data set, wherein the data cleaning network framework includes a range gate module and a feature classification module, and the input initial feature engineering set is processed by adopting a series or parallel network construction method to generate a target data set that filters out clutter points.

[0031] It should be noted that drones usually fly at low altitudes, and low-altitude interference sources are of various types and have different characteristics. The interference sources here include buildings, branches and leaves, birds, kites, balloons, etc. Due to differences in material, shape and movement characteristics, there are certain differences in radar echo signals.

[0032] The composition of the radar echo signal here usually includes: distance, azimuth, elevation, intensity, signal-to-noise ratio, filter number, base, elevation and difference ratio, azimuth and difference ratio, azimuth deviation, elevation intensity, azimuth intensity, elevation deviation, target original amplitude, target background amplitude, constant false alarm type, EP quality, etc. These signal indicators constitute an initial feature engineering set, and through technical methods and means, the main signals that reflect the differences in the detection objects can be screened out.

[0033] Data cleaning is a business process to remove clutter, and it is also a technical project for data preprocessing. Domestic and foreign experts have conducted a lot of research and exploration. Typical progress includes: clutter removal from the radar equipment side, suppressing clutter through algorithm design and application, such as spatial clutter suppression methods based on singular value decomposition, signal analysis methods based on multi-domain group images, etc.; clutter removal from the perspective of artificial neural network application, using fuzzy neural network, support vector machine technology, K-nearest neighbor algorithm and other methods and technologies, designing artificial neural network algorithms and classifiers to achieve clutter removal.

[0034] Aiming at the problem of radar echo signal data cleaning, this paper attempts to use artificial intelligence technology and traditional correlation gate methods, and innovates the design of generators, gradient optimization, feature engineering and other technical methods to form a flexible network, and studies the networking method algorithm to achieve the quality improvement of clutter signal removal. In order to achieve the goal of quality improvement, it is necessary to solve a series of problems such as the incompleteness of the current feature engineering method, the generation of the center origin of the correlation gate, the intermediate processing method, and the cross-fusion method.

[0035] This application proposes a data cleaning network framework for low-altitude UAV detection echo data, and further proposes specific implementation methods or algorithms from five aspects: network framework, wave gate design, feature classification design, intermediate processing, and cross fusion.

[0036] In view of the fact that both the current feature engineering method and the correlation gate method are difficult to obtain high-quality cleaning effects, this application proposes a data cleaning method that combines the use of distance gates and feature engineering. By cross-using the generator and discriminator in the field of artificial intelligence and adopting a series or parallel network construction method, a hybrid cleaning of multiple neural network models is achieved, thereby maximizing the target point screening effect of radar echoes. Figure 2 and Figure 3 ,in, Figure 2 A schematic diagram of a serial data cleaning network structure for drone swarm target detection provided in an embodiment of the present application; Figure 3 A schematic diagram of a parallel data cleaning network structure for drone swarm target detection provided in an embodiment of the present application.

[0037] It should be noted that in Figure 2 and 3In the application of neural network modules, the generator and gradient optimization modules are mainly used. The differences in the principles and characteristics of the two modules are as follows: the generator mainly builds the prediction ability of the neural network by learning the distribution characteristics of the data, generates new data, and combines the generated new data with the traditional correlation gate method to achieve the effect of filtering clutter points; the gradient optimization mainly selects several typical difference signals by learning and optimizing the echo feature set, and realizes the target point screening under the unsupervised learning method. The combined use of the two modules maximizes the advantages of their respective technologies and filters clutter signals.

[0038] There are some differences between the serial data cleaning network structure and the parallel data cleaning network structure, which are suitable for different scenarios and needs.

[0039] First, the structure of the serial data cleaning network is explained.

[0040] The characteristic of the serial type is that the original data set is processed twice independently in time series. The biggest feature here is that the output data quality of the range gate unit determines the input data quality of the feature classification unit, which directly affects the efficiency of feature classification. Therefore, it is necessary to use the intermediate processing unit to organize and optimize the output data, and adapt it to the high-quality input requirements of the feature classification unit while maintaining the authenticity of the data.

[0041] In an embodiment of the present application, the initial feature engineering set is processed by the data cleaning network framework to generate a target data set, including: inputting the initial feature engineering set into the range gate module of the serial data cleaning network framework, learning the initial feature engineering set through a generator, building the prediction ability of a neural network, generating a first data set, and combining the first data set with an associated gate to obtain a radar point trace prediction value, and performing signal screening based on the radar point trace prediction value to generate a second data set, wherein the range gate module includes a generator and an associated gate; inputting the generated second data set into an intermediate processing module for processing to generate a third data set; inputting the third data set into a feature classification module, screening through the feature engineering set, obtaining a fourth data set that meets the target batch signal, and inputting the obtained fourth data set into a gradient optimization module to screen out the difference signal between the target echo signal and the interference echo signal to generate a target data set.

[0042] It should be noted that the distance gate here includes two parts: a generator and an associated gate.

[0043] The generator mainly trains the artificial neural network model, simulates the distribution law of learning data, and builds the data prediction ability of the artificial neural network. The basic model can be a standard machine learning model, a deep learning model, or a combination of various models or a new neural network with added auxiliary processing capabilities.

[0044] The correlation gate mainly draws on the traditional radar detection correlation gate technology method, with the distance gate as the main method. Based on the generator, it obtains the prediction value of the next step or several steps, that is, the radar point trace prediction value. With the prediction value as the origin, a suitable distance threshold is set to achieve the screening of echo signals.

[0045] The feature classification module here includes two parts: feature engineering set and gradient optimization.

[0046] The feature engineering set refers to the signal set in the radar echo signal that characterizes the distinction between the target echo and clutter. Usually, candidate echo signal indicators include: target original amplitude, target background amplitude, Doppler velocity, filter flag, constant false alarm type, noise-to-clutter ratio, EP quality, etc.

[0047] The target echo and clutter show a certain degree of difference in candidate signals, such as inconsistent range, different type, different value, etc. However, in different operating environments, due to differences in interference sources in the airspace background, different shapes and materials of the detected targets, even the same signal indicators will have different performances. Therefore, it is necessary to optimize the feature engineering set to find the best indicator combination that meets the current batch of signals.

[0048] Gradient optimization refers to the use of neural network models to convert the optimization problem of feature engineering sets into a global optimization problem of neural network models through gradient optimization design and learning training. Through gradient optimization, several signals that best reflect the difference between target echo and clutter can be selected from the feature engineering set.

[0049] The intermediate processing module and algorithm here mainly perform a series of quality enhancement operations on the data generated by the range gate, including signal missing blind spot filling, data expansion caused by state changes, data compression of formation targets, and data expansion of single equipment.

[0050] Specifically, the second data set is input into the intermediate processing module for processing to generate the third data set, including: in response to detecting that there is missing signal data in the second data set, the second data set is point-filled; that is, the missing signal is filled, and the data is point-filled for the period when the radar detection echo signal is briefly interrupted and the dot disappears. The common reason for the interruption of the echo signal is usually the obstruction of obstacles.

[0051] In response to detecting a change in the target motion state data in the second data set, the data search range of the second data set is expanded; that is, the state change, in order to detect the change in the motion state of the target, such as the adjustment from straight-line motion to turning motion, the randomness of the echo signal caused by the motion is increased and the timeliness of the tracking algorithm is insufficient, which reduces the reliability of the data. In this case, the range of the data output by the range gate needs to be expanded to enhance the coverage effectiveness of the data.

[0052] In response to detecting the presence of a formation target in the second data set, the peripheral data other than the formation target data is eliminated; that is, the formation target. In the case of detecting a formation of targets, since the echo signal presents multiple tracks, the echo stability of the track is more obvious, therefore, the output data quality of the range gate is better, and the peripheral data can be cropped.

[0053] In response to detecting the presence of a single target in the second data set, the data boundary of the second data set is expanded; that is, the single target. In contrast to the formation target, the single target has a single signal and increased randomness, which reduces the reliability of the data. It is necessary to expand the data boundary to ensure good coverage.

[0054] The third data set is generated by performing the point completion operation, and / or the data search range expansion operation, and / or the peripheral data elimination operation, and / or the data boundary expansion operation.

[0055] More specifically, in response to detecting that there is signal missing data in the second data set, performing point completion on the second data set includes:

[0056] When the signal density of the target point trace coverage area is less than a preset density threshold, it is determined that there is signal missing data in the second data set, and a point trace completion operation is triggered;

[0057] When the signal density in the target point trace coverage area is not less than a preset density threshold, it is determined that the signal in the second data set is normally covered.

[0058] In practical applications, if the signal density of a certain trace coverage area is lower than 50% of the normal density, it is considered that there is a trace interruption and the completion operation is triggered; if it is greater than 50%, it is considered to be normal coverage.

[0059]

[0060] Among them, δ is the trigger parameter and ρ is the coverage density.

[0061] The completion operation mainly involves the use of the generator to generate data covering the specified time period area.

[0062] More specifically, in response to detecting that there is a change in the target motion state data in the second data set, expanding the data search range of the second data set includes:

[0063] Taking the target base point as the starting point, the direction of the line segment for three consecutive steps after the starting point is obtained. When the deviation angle of the direction of the line segment for three consecutive steps after the starting point meets the preset deviation angle condition, it is determined that there is a change in the target motion state data in the second data set, which triggers the operation of expanding the data search range.

[0064] In practical applications,

[0065] Using a three-step measurement algorithm, starting from a certain base point, measure the straight line direction of the next three steps. If the deviation angle continues to increase, it is considered a turning motion. Set the plane coordinates of the base point (x0, y0), and the coordinates of the next three steps (x1, y1), (x2, y2), (x3, y3).

[0066]

[0067] Among them, α is the trigger parameter, θ1, θ2, θ3 are the direction angles. The operation is mainly to expand the data boundary, that is, to restore part of the data filtered out by the range gate.

[0068] More specifically, in response to detecting that a formation target exists in the second data set, removing peripheral data other than the formation target data includes:

[0069] When there are formation targets consisting of more than three targets in the second data set, an operation of removing surrounding data is triggered;

[0070] When there are two targets in the second data set forming a formation target, the data in the second data set remains unchanged.

[0071] In practical applications, for formation targets and single targets, the specific number of targets is determined. When the number of targets is greater than 3, data is trimmed; when the number of targets is 2, it remains unchanged; when the number of targets is 1, data is expanded.

[0072] Then the structure of the parallel data cleaning network is explained.

[0073] For parallel networks, since the feature classification units of the range gate units output data simultaneously and converge to form the final target data, there is a certain degree of filtering overlap in their respective processing processes. Therefore, it is necessary to combine application requirements and adopt appropriate cross-fusion methods to improve the effectiveness of output data.

[0074] In the embodiment of the present application, 7. The method according to claim 2 is characterized in that the processing of the initial feature engineering set by a data cleaning network framework to generate a target data set includes:

[0075] The initial feature engineering set is input into the wave gate module and the feature classification module at the same time to generate a fifth data set and a sixth data set, wherein the fifth data set is a data set generated after the initial feature engineering set is processed by the generator and the associated wave gate in sequence; the sixth data set is a data set generated after the initial feature engineering set is processed by the feature engineering set and the gradient optimization module in sequence;

[0076] The fifth data set and the sixth data set are fused by a cross fusion algorithm to generate a target data set.

[0077] It should be noted that the cross-fusion algorithm here mainly designs a trust mechanism for the two types of outputs, namely, the range gate and feature classification, to determine the reliability of the data, and further provides a method for generating synthetic data. Factors affecting the trust include: the number of detected targets, the material of the detected targets, the complexity of the detection background environment, etc.

[0078] The number of detected targets affects the strength of the target signal. Multi-target signals can enhance the distinction from clutter. The material of the detected target affects the numerical difference of the specific category of the echo signal. The detection background environment affects the threshold of the range gate, etc. A trust characterization method is proposed. Let the trust index of the range gate be ε, the generated data set be A, the feature classification trust index be γ, and the output data set be B. Then the union-based fusion is as follows: C = A∪B; the intersection-based fusion is as follows: C = A∩B.

[0079] Specifically, the step of fusing the fifth data set and the sixth data set by a cross fusion algorithm to generate a target data set includes:

[0080] Based on a target cross-fusion calculation formula, the fifth data set and the sixth data set are fused to generate a target data set, wherein the target cross-fusion calculation formula includes:

[0081] C=(rand()*ε*(AA∩B))∪(A∩B)∪(rand()*γ*(BA∩B))

[0082] Among them, C is the target data set after fusion; rand() is a random seed, which means that a certain proportion of data is randomly selected from the data complement outside A∩B according to the trust parameter; ε is the distance gate trust index; γ is the feature classification trust index; A is the fifth data set; B is the sixth data set.

[0083] The data cleaning method proposed in the embodiment of the present application is suitable for detection scenarios of drones or clusters in low-altitude environments, such as low-altitude airspace in cities, border plateaus and mountains, and low-altitude airspace near the sea. Usually, due to the large number of interference sources contained in the operation scene, the quality of the original data of the detection echo is poor, and effective technical means are required for cleaning operations. The proposed neural network model application system and implementation algorithm can efficiently remove clutter and obtain high-quality data samples.

[0084] To apply the method provided in this application, the following aspects need to be done: the first is to do a good job in feature engineering design. According to the experience of signal experts in the radar field, the initial feature engineering set under different scenarios is selected to maximize the difference in signal characteristics; the second is to do a good job in model design. Due to differences in implementation algorithms, the effects of series networks and parallel networks in different scenarios will be different. It is necessary to do a good job in model design, improve the quality of model fitting, and cooperate with the network structure to achieve better data cleaning effects; the third is to do a good job in the specific design of intermediate processing algorithms and cross-fusion algorithms, so that the effectiveness of distance gates and feature classification can be maximized.

[0085] By applying the data cleaning method for drone swarm target detection provided in the embodiment of the present application, the collected initial feature engineering set is processed for clutter removal through a data cleaning network framework, which can improve the clutter removal efficiency and generate a high-quality target data set.

[0086] Corresponding to the above method embodiment, this specification also provides an embodiment of a data cleaning device for drone group target detection, Figure 4 A schematic diagram of the structure of a data cleaning device for drone group target detection provided by an embodiment of the present application. Figure 4 As shown, the device comprises:

[0087] The acquisition module 402 is configured to acquire radar echo signals generated by the UAV in the target area under the influence of the interference source to form an initial feature engineering set, wherein the radar echo signals include target echo signals and interference echo signals;

[0088] The processing module 404 is configured to process the initial feature engineering set through a data cleaning network framework to generate a target data set, wherein the data cleaning network framework includes a range gate module and a feature classification module, and processes the input initial feature engineering set by adopting a series or parallel network construction method to generate a target data set that filters out clutter points.

[0089] In an optional embodiment, the processing module 404 is further configured to:

[0090] The initial feature engineering set is input into the range gate module of the series data cleaning network framework, the initial feature engineering set is learned by the generator, the prediction capability of the neural network is constructed, a first data set is generated, and the first data set is combined with the correlation gate to obtain the radar point trace prediction value, and based on the radar point trace prediction value, signal screening is performed to generate a second data set, wherein the range gate module includes a generator and an association gate;

[0091] Inputting the generated second data set into an intermediate processing module for processing to generate a third data set;

[0092] The third data set is input into a feature classification module, and is screened through a feature engineering set to obtain a fourth data set that meets the target batch signal. The fourth data set obtained is input into a gradient optimization module to screen out the distinguishing signals between the target echo signal and the interference echo signal, thereby generating a target data set.

[0093] In an optional embodiment, the processing module 404 is further configured to:

[0094] In response to detecting that there is signal missing data in the second data set, performing point completion on the second data set;

[0095] In response to detecting a change in the target motion state data in the second data set, expanding a data search range of the second data set;

[0096] In response to detecting that a formation target exists in the second data set, removing peripheral data other than the formation target data;

[0097] In response to detecting that a single object exists in the second data set, expanding a data boundary of the second data set;

[0098] The third data set is generated by performing the point completion operation, and / or the data search range expansion operation, and / or the peripheral data elimination operation, and / or the data boundary expansion operation.

[0099] In an optional embodiment, the processing module 404 is further configured to:

[0100] When the signal density of the target point trace coverage area is less than a preset density threshold, it is determined that there is signal missing data in the second data set, and a point trace completion operation is triggered;

[0101] When the signal density in the target point trace coverage area is not less than a preset density threshold, it is determined that the signal in the second data set is normally covered.

[0102] In an optional embodiment, the processing module 404 is further configured to:

[0103] Taking the target base point as the starting point, the direction of the line segment for three consecutive steps after the starting point is obtained. When the deviation angle of the direction of the line segment for three consecutive steps after the starting point meets the preset deviation angle condition, it is determined that there is a change in the target motion state data in the second data set, which triggers the operation of expanding the data search range.

[0104] In an optional embodiment, the processing module 404 is further configured to:

[0105] When there are formation targets consisting of more than three targets in the second data set, an operation of removing surrounding data is triggered;

[0106] When there are two targets in the second data set forming a formation target, the data in the second data set remains unchanged.

[0107] In an optional embodiment, the processing module 404 is further configured to:

[0108] The initial feature engineering set is input into the wave gate module and the feature classification module at the same time to generate a fifth data set and a sixth data set, wherein the fifth data set is a data set generated after the initial feature engineering set is processed by the generator and the associated wave gate in sequence; the sixth data set is a data set generated after the initial feature engineering set is processed by the feature engineering set and the gradient optimization module in sequence;

[0109] The fifth data set and the sixth data set are fused by a cross fusion algorithm to generate a target data set.

[0110] In an optional embodiment, the processing module 404 is further configured to:

[0111] Based on a target cross-fusion calculation formula, the fifth data set and the sixth data set are fused to generate a target data set, wherein the target cross-fusion calculation formula includes:

[0112] C=(rand()*ε*(AA∩B))∪(A∩B)∪(rand()*γ*(BA∩B))

[0113] Among them, C is the target data set after fusion; rand() is a random seed, which means that a certain proportion of data is randomly selected from the data complement outside A∩B according to the trust parameter; ε is the distance gate trust index; γ is the feature classification trust index; A is the fifth data set; B is the sixth data set.

[0114] By using the data cleaning device for drone swarm target detection provided in the embodiment of the present application, the collected initial feature engineering set is processed for clutter removal through a data cleaning network framework, which can improve the clutter removal efficiency and generate a high-quality target data set.

[0115] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the data cleaning device for drone swarm target detection, since it is basically similar to the data cleaning method embodiment for drone swarm target detection, the description is relatively simple, and the relevant parts can be referred to the partial description of the data cleaning method embodiment for drone swarm target detection.

[0116] Figure 5 This is a block diagram of a computing device provided in an embodiment of the present application. The components of the computing device 500 include but are not limited to a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and the database 550 is used to store data.

[0117] The computing device 500 also includes an access device 540 that enables the computing device 500 to communicate via one or more networks 560. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of network interface (e.g., a network interface card (NIC)) that is wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a USB (Universal Serial Bus) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

[0118] In one embodiment of the present specification, the above components of the computing device 500 and Figure 5 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 5 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0119] The computing device 500 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 500 may also be a mobile or stationary server.

[0120] Among them, the processor 520 is used to execute the following computer executable instructions, which, when executed by the processor, implement the steps of the data cleaning method for drone swarm target detection.

[0121] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computing device embodiment, since it is basically similar to the data cleaning method embodiment for drone swarm target detection, the description is relatively simple, and the relevant parts can be referred to the partial description of the data cleaning method embodiment for drone swarm target detection.

[0122] An embodiment of the present specification also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the data cleaning method for drone swarm target detection.

[0123] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computer-readable storage medium embodiment, since it is basically similar to the data cleaning method embodiment for drone swarm target detection, the description is relatively simple, and the relevant parts can be referred to the partial description of the data cleaning method embodiment for drone swarm target detection.

[0124] An embodiment of the present specification also provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned data cleaning method for drone swarm target detection.

[0125] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computer program embodiment, since it is basically similar to the data cleaning method embodiment for drone swarm target detection, the description is relatively simple, and the relevant parts can be referred to the partial description of the data cleaning method embodiment for drone swarm target detection.

[0126] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0127] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0128] It should be noted that the above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of the present specification.

[0129] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0130] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A data cleaning method for drone group target detection, characterized in that: include: Collect radar echo signals generated by UAVs in the target area under the influence of interference sources to form an initial feature engineering set, wherein the radar echo signals include target echo signals and interference echo signals; The initial feature engineering set is processed by a data cleaning network framework to generate a target data set, wherein the data cleaning network framework includes a range gate module and a feature classification module, and processes the input initial feature engineering set by adopting a series or parallel network construction method to generate a target data set that filters clutter points.

2. The method according to claim 1, characterized in that The data cleaning network framework includes a series data cleaning network framework and a parallel data cleaning network framework. The processing of the initial feature engineering set by a data cleaning network framework to generate a target data set includes: The initial feature engineering set is input into the range gate module of the series data cleaning network framework, the initial feature engineering set is learned by the generator, the prediction capability of the neural network is constructed, a first data set is generated, and the first data set is combined with the correlation gate to obtain the radar point trace prediction value, and based on the radar point trace prediction value, signal screening is performed to generate a second data set, wherein the range gate module includes a generator and an association gate; Inputting the generated second data set into an intermediate processing module for processing to generate a third data set; The third data set is input into a feature classification module, and is screened through a feature engineering set to obtain a fourth data set that meets the target batch signal. The fourth data set obtained is input into a gradient optimization module to screen out the distinguishing signals between the target echo signal and the interference echo signal, thereby generating a target data set.

3. The method according to claim 2, characterized in that The step of inputting the generated second data set into an intermediate processing module for processing to generate a third data set includes: In response to detecting that there is signal missing data in the second data set, performing point completion on the second data set; In response to detecting a change in the target motion state data in the second data set, expanding a data search range of the second data set; In response to detecting that a formation target exists in the second data set, removing peripheral data other than the formation target data; In response to detecting that a single object exists in the second data set, expanding a data boundary of the second data set; The third data set is generated by performing the point completion operation, and / or the data search range expansion operation, and / or the peripheral data elimination operation, and / or the data boundary expansion operation.

4. The method according to claim 3, characterized in that In response to detecting that there is signal missing data in the second data set, performing point completion on the second data set includes: When the signal density of the target point trace coverage area is less than a preset density threshold, it is determined that there is signal missing data in the second data set, and a point trace completion operation is triggered; When the signal density in the target point trace coverage area is not less than a preset density threshold, it is determined that the signal in the second data set is normally covered.

5. The method according to claim 3, characterized in that: In response to detecting that there is a change in the target motion state data in the second data set, expanding the data search range of the second data set includes: Taking the target base point as the starting point, the direction of the line segment for three consecutive steps after the starting point is obtained. When the deviation angle of the direction of the line segment for three consecutive steps after the starting point meets the preset deviation angle condition, it is determined that there is a change in the target motion state data in the second data set, which triggers the operation of expanding the data search range.

6. The method according to claim 3, characterized in that In response to detecting that there is a formation target in the second data set, removing peripheral data other than the formation target data includes: When there are formation targets consisting of more than three targets in the second data set, an operation of removing surrounding data is triggered; When there are two targets in the second data set forming a formation target, the data in the second data set remains unchanged.

7. The method according to claim 2, characterized in that The initial feature engineering set is processed by the data cleaning network framework to generate a target data set, including: The initial feature engineering set is input into the wave gate module and the feature classification module at the same time to generate a fifth data set and a sixth data set, wherein the fifth data set is a data set generated after the initial feature engineering set is processed by the generator and the associated wave gate in sequence; the sixth data set is a data set generated after the initial feature engineering set is processed by the feature engineering set and the gradient optimization module in sequence; The fifth data set and the sixth data set are fused by a cross fusion algorithm to generate a target data set.

8. The method according to claim 7, characterized in that The step of fusing the fifth data set and the sixth data set by a cross-fusion algorithm to generate a target data set includes: Based on a target cross-fusion calculation formula, the fifth data set and the sixth data set are fused to generate a target data set, wherein the target cross-fusion calculation formula includes: C=(rand()*ε*(AA∩B))∪(A∩B)∪(rand()*γ*(BA∩B)) Among them, C is the target data set after fusion; rand() is a random seed, which means that a certain proportion of data is randomly selected from the data complement outside A∩B according to the trust parameter; ε is the distance gate trust index; γ is the feature classification trust index; A is the fifth data set; B is the sixth data set.

9. A data cleaning device for drone group target detection, characterized in that: include: A collection module is configured to collect radar echo signals generated by the UAV in the target area under the influence of the interference source to form an initial feature engineering set, wherein the radar echo signals include target echo signals and interference echo signals; The processing module is configured to process the initial feature engineering set through a data cleaning network framework to generate a target data set, wherein the data cleaning network framework includes a range gate module and a feature classification module, and processes the input initial feature engineering set by adopting a series or parallel network construction method to generate a target data set that filters out clutter points.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an implementation program for information transmission, and when the program is executed by the processor, the steps of the data cleaning method for drone swarm target detection described in any one of claims 1 to 8 are implemented.

Images